Scholarlink

<p>Proper and precise observation of water fractions is essential to control food drying process, and ensure product quality along efficient energy consumption. Current research has determined the potential of ensemble machine learning algorithms to predict water fraction status during freeze-drying of mushroom using near-infrared spectral range. Five ensemble machine learning (ML) models like random forest (RF), gradient boosting (GB), extreme gradient boosting (XGBoost), CatBoost and LightGBM regressions have been evaluated and compared. Performance accuracy was assessed using the coefficient of determination (R²), root mean square error (RMSE), relative root mean square error (RRMSE), bias and residual predictive deviation (RPD). Among these non-optimized ensemble models, XGBoost obtained the highest values of prediction accuracy metrics R²p for total water (TW) of 0.8256. Further analysis revealed that GB and XGBoost acquired comparable accuracy coefficient of determination for free water (FW) and bound water (BW) of 0.8400 and 0.9580 respectively. Similar trend in error levels is also observed for both of these models. In terms of robustness, RPD further demonstrated fraction specified model strength as RF has shown superior RPD for total water, XGBoost for FW and IW, while GB obtained highest score in BW. These findings indicate that model robustness is water fraction dependent instead of focusing by a single algorithm. Performance indicating that these approaches are effective in capturing highly variable nonlinear complex spectra.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Comparative assessment of ensemble machine learning models for detecting water dynamics in freeze-dried mushrooms using Vis-NIR spectroscopy

  • 期刊论文

  • 发表日期：

    2026年1月30日

查看全文
  • Shoaib Younas,
  • Farhan Ali,
  • Ukasha Arqam,
  • Xin Wang,
  • Muhammad Ahmad,
  • Fatima Tariq,
  • Kashaf ul Eman,
  • Hafiz Rizwan Sharif,
  • Zeshan Ali

摘要

Proper and precise observation of water fractions is essential to control food drying process, and ensure product quality along efficient energy consumption. Current research has determined the potential of ensemble machine learning algorithms to predict water fraction status during freeze-drying of mushroom using near-infrared spectral range. Five ensemble machine learning (ML) models like random forest (RF), gradient boosting (GB), extreme gradient boosting (XGBoost), CatBoost and LightGBM regressions have been evaluated and compared. Performance accuracy was assessed using the coefficient of determination (R²), root mean square error (RMSE), relative root mean square error (RRMSE), bias and residual predictive deviation (RPD). Among these non-optimized ensemble models, XGBoost obtained the highest values of prediction accuracy metrics R²p for total water (TW) of 0.8256. Further analysis revealed that GB and XGBoost acquired comparable accuracy coefficient of determination for free water (FW) and bound water (BW) of 0.8400 and 0.9580 respectively. Similar trend in error levels is also observed for both of these models. In terms of robustness, RPD further demonstrated fraction specified model strength as RF has shown superior RPD for total water, XGBoost for FW and IW, while GB obtained highest score in BW. These findings indicate that model robustness is water fraction dependent instead of focusing by a single algorithm. Performance indicating that these approaches are effective in capturing highly variable nonlinear complex spectra.